Four diploid Saccharomyces cerevisiae strains named L71, L69, L62, and L60 were provided by Lallemand Inc. (Montréal, Canada). The strain Saccharomyces cerevisiae BY4743 (MATa/α his3Δ1/his3Δ1 leu2Δ0/leu2Δ0 LYS2/lys2Δ0 met15Δ0/MET15 ura3Δ0/ura3Δ) was obtained from Euroscarf collection. Unless otherwise stated, yeast cells were cultivated at 30°C in 1 L Erlenmeyer flask containing a 200 ml of YPD medium (2% w/v glucose, 1% w/v peptone, 1% w/v yeast extract) in a rotary shaker set at 200 rpm.

At least three independent cultures were carried out in 200 ml of YPD as described above. Cells were collected during the exponential phase at OD₆₀₀ of around 1.0 (correspond to 2 × 10⁷ cells/ml). Cells were harvested by centrifugation 10 min at 3,000 g and washed two times with sterile water. The pellets were kept to isolate cell walls. Cell walls were isolated and purified by centrifugation and extensive washing as described in Francois (2006). Polysaccharides mannans, chitin, β-1,3- and β-1,6-glucans in the purified cell walls were determined as described in Schiavone et al. (2014). Quantification of the released monomers (mannose, glucose and N-acetylglucosamine) was determined by high performance anionic exchange chromatography (HPAEC) coupled to amperometric detection as described by Dallies et al. (1998), and colorimetric method, respectively (Reissig et al., 1955).

Hydrophobic properties of yeasts strains were determined by measuring their affinity for an apolar solvent as described in Purevdorj-Gage et al. (2007). Briefly, overnight cultures were centrifuged at 2,000 g for 5 min and resuspended in fresh YPD medium to obtain an optical density of 1 at 600 nm. After 3 h of incubation at room temperature, the OD₆₀₀ was measured for each culture. In 15 × 100 mm borosilicate glass tubes 0.6 ml of octane (Sigma-Aldrich) was added to 1.2 ml of yeast cell suspension. The mixtures were vortexed for 120 s and allowed to stand for 15 min at room temperature to achieve the complete separation of the two phases. The aqueous phase was recovered and its OD₆₀₀ measured. The results were expressed as the Octane adhesion index, (% A) which represents the percentage of cells retained by the organic fraction, according to the relationship: % A = [A₀-A_F/A_F] × 100. A₀ and A_F represent the optical density at 600 nm of the yeast suspension before and after contact with octane. Assays were performed in triplicates.

For each yeast strains, three independent cultures were carried out in 50 ml of YPD in a 250 mL shake flasks on a rotary shaker set at 200 rpm at 30°C. Yeast cells (a total of about 10 OD₆₀₀ units) were collected at OD₆₀₀ at about 1.0 by centrifugation (3,000 rpm, 4°C, 2 min), followed by a washing step with 1 mL of sterilized water. The cell pellets were immediately frozen in liquid nitrogen and stored at −80°C until RNA extraction.

The package mixOmics employs partial least square (PLS) based methods to unravel relationships between variables from multiple biological datasets. The tool has its own tutorial and can be easily used, for a nearly inexperienced R user, through the web site at http://mixomics.org/. Specific functions were searched on the Saccharomyces Genome Database (http://www.yeastgenome.org), Yeastract (http://www.yeastract.com/) and Funspec (http://funspec.med.utoronto.ca/).

Biochemical composition of cell wall was determined in the four industrial and the laboratory strains cultivated that were harvested in exponential phase (i.e., OD₆₀₀ ~1 unit) of growth on a glucose rich medium. Table 1 recapitulates the data obtained for the three polysaccharides that constitute the yeast cell wall. It can be see that the mannans content in cell wall of three out of the 4 industrial strains was 30% higher than in BY4743 strain (laboratory strain). This higher mannans content in strains L60 and L69 was actually expected since these strains are wine yeasts, whereas L20 has been selected after mutagenesis for higher mannoproteins content (Lallemand Inc., personal communication). In correlation with this high content of mannans, the hydrophobicity of these three strains was 3–6-times higher than that of the lab strain. However, this property is likely not entirely associated with levels of mannans since the industrial strain L71 exhibited a similar hydrophobicity as L62 and L60 strain albeit the proportion of mannans in L71 strain was comparable to that of the laboratory strain BY4743. An additional relevant data that reveals strain-dependency of the cell wall composition is concerning the β-glucans content and more specifically the β-1,3-glucans/β-1,6-glucans ratio. In the industrial strains L71, L62, and L60, the proportion of β-1,6-glucans reached more than 50% of total β-glucans, whereas it represented less than 30% in the lab strain, in accordance with a previous report (Schiavone et al., 2014). Whether this higher proportion of β-1,6-glucans in industrial strains in spite of lower amount of total β-glucan in these strains as compared to the lab strain can explain the remarkable efficacy of these industrial strains to stimulate the immune response (Lallemand Inc., unpublished data) remains to be verified. Finally, the proportion of chitin in cell wall was found around 4% in all strains, which agrees with previous works (Schiavone et al., 2014), with the exception of L62 strain for which it was 2 times higher. Collectively, these results showed high variability in the composition of the cell between strains, which may result from the selection and adaptation of yeasts to specific cultures/process conditions.

To investigate on a potential association between the biochemical composition and the nanomechanics of the cell wall, we determined the elasticity (Young) modulus by performing nano-indentations experiments in which the deformation depth is plotted as a function of the force applied by the AFM silicon nitride tip on the cell. For these AFM experiments, the yeast cells were individually trapped in PDMS-fabricated microchambers according to the method described previously (Dague et al., 2011). AFM imaging of one individual cell from each strain is illustrated in Figure 1A and shows relatively identical surface topology of these yeasts. For elasticity modulus, the AFM measurements were performed on areas devoid of bud scars and on at least 5 independent cells for statistical analysis of the data. As described in section Materiel and Methods, 1,024 indentation curves are performed on each cell and fitted to the Hertz model. This generates histograms of Young's modulus value that can be adjusted to a Gaussian distribution The most probable value of the Young's modulus corresponding to the maximal height of the Gaussian curve gives a quantitative indication of the cell stiffness. It is shown in Figure 1B that this value is remarkably different between strains and hardly difficult to correlate with the biochemical composition of the cell wall (Table 1). Notably, strain L62 which contains twice more chitin in the cell wall than the four other strains exhibited the lowest Young's modulus value of 239 kPa, whereas it was previously reported that the chitin level is important for the stiffness of the cells (Touhami et al., 2003). On the other hand, the industrial strain L60 and lab strain have a comparable elasticity modulus, but the content in β-glucans and mannans in these two strains is completely different. Taken together, these data reinforce our earlier work showing that the elasticity property of a yeast cell is independent on its cell wall composition (Dague et al., 2010; Francois et al., 2013).

The difference in hydrophobicity between industrial and lab strains which is at first glance not directly associated with higher levels of mannans is suggestive of some dissimilarity in the organization, structure or degree of mannosylation of cell wall proteins between these strains. To investigate this question, we employed single molecule force spectroscopy (SMFS) using AFM-tips functionalized with concanavalin A (AFM tip-ConA), a plant lectin known to bind specifically α-mannose residues of glycoproteins (So and Goldstein, 1968; Gad et al., 1997). This technique allows to probe the distribution, adhesion, flexibility and extension of mannoproteins at the surface of yeast cells. Figure 2A shows that the surface topology of these strains as recorded in Quantitative Imaging™ mode (Chopinet et al., 2013) was very comparable. However, the adhesion frequency of functionalized tip at the cell surface vs. the force needed to break the interaction (unbinding force) was only 10% for strain BY while these adhesion events were in the range of 25–35% with these industrial strains. Thus, these data indicated that the cell surface of industrial strain is physically distinct from that of the lab strain. In spite of this difference, the adhesion force which also corresponds to the force needed to unbind the interaction of the tip-ConA with the cell surface was estimated in the range of 55–60 pN (Figure 2B). This value is very similar to the one obtained in previous works (Alsteens et al., 2008; Francius et al., 2009) using AFM tip functionalized with ConA through a 6 nm long polyethylene glycol (PEG) chain. This data indicates that the unbinding force for a single lectin—α-mannose interaction is not modified by the type of spacer between the tip and the ConA protein. Finally, we confirmed that the AFM-tip ConA directly interacts with cell wall mannoproteins since the interaction was almost complete lost by adding 100 mM mannose prior to the AFM measurements (data not shown). Figure 2C also shows that single adhesion events take place as deduced from the pattern of force-distance curves obtained with AFM-tip ConA, except for L69 strain which revealed more than one event. Therefore, for strains BY4743, L71, L62, L60, the unbinding force of the single adhesion event could fit a freely jointed chain (FJC) model with a Kuhn length (k_l) in the range of 0.05–1 nm (see Figure S1). For L69 strain, the multiple unbinding peaks were well fitted with a WLC model with persistent lengths (Ip) ranging from 0.025 to 2 nm. This high variation of this parameter suggested a relative high flexibility of the mannans chains or mannoproteins pulled by the AFM tip.

Use of the FJC and WLC models allowed to determine the contour length, which correspond to the extension or relaxation of mannans chains obtained by pulling with the AFM tip-ConA at low force until the force necessary to extend further rises rapidly and ends up by breaking the binding (Fisher et al., 1999). As illustrated in Figure 3A, the contour length values roughly followed a Gaussian distribution, with the shape of the curve and the mean value not the same between the strains, suggesting some structural differences of the cell wall mannans/mannoproteins in these strains. Except with strain 69 for which the contour length distribution ranged from 20 to 300 nm, with a mean value around 97 nm, the distribution was sharp with a mean value centered at 20–25 nm for strain L71 and L60, and was 2.5 times higher in strains L62 and BY4743. An additional pertinent physical data that can be obtained from the adhesion force vs. distance curves is the rupture distance (Figure 3B) which corresponds to the distance at which the interaction of ConA with the mannosyl unit of the mannoproteins is broken, i.e., when the adhesion force is reset to zero (see Figure S1). Interestingly, strain L69 that showed the longest contour length also presented the longest rupture distances that could reach up to 400 nm. However, this relationship between contour length and rupture distance was not respected in the other strains since the rupture distances in strain 71 were about 2 times higher than in strain 60 albeit both strains exhibited similar contour lengths (Figure 3B). A similar statement could be made for strain BY4743 and L62. Altogether, these data highlighted significant differences in the physical properties of cell wall mannans between yeast strains.

The biochemical and biophysical data reported above raised the question whether these differences can be in part related to changes in genes expression between strains. We addressed this question by exploring and comparing the transcriptome of industrial and laboratory strains using DNA microarrays technology. As a first step, we analyzed the global transcriptomic data from three biological replicates of each strain using Partial least square (PLS) regression analysis developed elsewhere (Le Cao et al., 2008) as this method can highlight similarities and dissimilarities between samples. This analysis clearly denoted the good repeatability of the transcriptome analysis for each strain (Figure S2). It also illustrated that the transcriptomes of the 4 industrial strains are relatively close and in particular those of strains L60 and L62 which are very neighbors, whereas the transcriptome of the diploid lab strain significantly separates from the four industrial yeasts. Looking for genes that were differentially expressed in the industrial strains relative to the lab strain BY4743, lists of 162, 210, 190, and 166 genes in L71, L69, L62, and L60 were retained, taking as criteria only genes that were differentially expressed by a factor ≥2 at a p-value < 0.01. Venn representation of these data showed that 10–25% of the differentially expressed genes were strain-specific and among these lists, a set of 71 genes emerged as being differentially expressed in all four industrial strains relative to the lab strain (Figure S3 and Table S1 for detailed description of the gene function).

Functional classification of these 71 genes into GO-molecular function and GO-cellular component showed enrichment of cell wall-bounded enzymes which included by genes encoding acid phosphatases and cell wall associated L-asparaginases and in mitochondrial function including heme binding and electron transport activity (Figure S4). A 2D-cluster analysis of this genes list could be further broken down into 5 subgroups (Figures 4A–E). The subgroup A included only 4 genes (YLR155C/ASP3-1, YLR157C/ASP3-2, YLR158C/ASP3-3, and YLR160C/ASP3-4) that encode the cell-wall or periplasmic localized L-asparaginases. The strong downregulation of ASP genes family in these four industrial strains are in line with previous results indicating their absence in the genome of 128 fungal species including wine and brewing yeasts (Pope et al., 2007; Carreto et al., 2008). Conversely, the subgroup E comprises LEU2, HIS3, and URA3 whose expression is extremely high in industrial yeasts because these 3 genes are used as auxotrophic marker in the laboratory strain and thus are transcriptionally defective. A large subgroup (B) of about 45 genes showed a slight (1.5) to moderate (3–4-fold) lower expression in industrial strains as compared to laboratory strain. Main GO biological processes in this group were related to phosphate metabolic process such as YHR215w/PHO12, YAR071w/PHO11, YDR281c/PHM6, and YERO57w/PHM8 whose expression is known to be induced by low P_i or repressed by high P_i (Ogawa et al., 2000). This data suggests that industrial strains may have a higher bioavailability of inorganic phosphate. This inferred “higher Pi content” in these strains is further illustrated in the subgroup C in which YBR093C/PHO5 encoding the major phosphate repressible acid phosphatase (Huang and O'Shea, 2005), the high affinity Na+/Pi cotransporter encoded PHO89/ YBR296c gene (Auesukaree et al., 2003) and SLP2 encoding a protein that targets the low affinity H⁺/Pi transporter encoded by PHO87 into vacuole for degradation (Ghillebert et al., 2011) were 10–20-fold less expressed than in the laboratory strain. Finally, the subgroup D comprised a short list of 12 slightly to moderately upregulated genes in industrial strains (i.e., from 1.5 to 4-fold more expressed) whose mitochondrial electron transport chain and mitochondrial respiratory complex III were identified as the main GO biological process and GO-component, respectively by the Funspec (http://funspec.med.utoronto.ca/) and GO term finder tools (http://www.yeastgenome.org/cgi-bin/GO/goTermFinder.pl). More specifically, upregulation of the RIP1 and CYT1 encoding two of the three catalytic subunits of the cytochrome bc1 and of CYC1 encoding cytochrome C1 suggests that, even under high glucose condition for which respiration is strongly repressed but very likely not completely shut off (Lagunas, 1986), the industrial strains may acquire more energy from the oxidative phosphorylation than the lab strain since cytochrome bc1 is critically important for this function (Hunte et al., 2003). Alternatively, the apparent higher respiratory activity of the industrial strains could be an indirect effect of the well-known poor respiratory competency of the laboratory strain BY4743 that is derived from S288c background (Young and Court, 2008).

The transcriptome profile of the 4 industrial yeast strains compared to that of the laboratory strain BY4743 revealed a remarkable enrichment of genes implicated in the cell wall architecture and remodeling. Thus, a total of 80 genes differentially expressed between industrial and laboratory strains was retrieved and subjected to a detailed functional analysis (see Table S2 for detailed description of the gene function). A 2D-clustering representation of these differentially expressed genes showed that, apart from the ASP genes family which are strongly downregulated likely because they are absent in these industrial strains (see above), most of the other cell wall related genes were broadly upregulated but to different extent in the 4 industrial strains (Figure 5 and see Figure S5 for G0 biological and molecular function classification). In particular, we noticed a potent upregulation of the adhesin-encoded FLO11/YIR019C and of YHR213w which codes for a protein sharing sequence similarity with the adhesin-encoded FLO1 gene (Teunissen and Steensma, 1995) in strain L69 and L60. Interestingly, this upregulation of these two genes which was the highest in strain L69 was concomitant with the highest hydrophobicity displayed by this strain (Table 1) and with the observation that this strain formed large aggregates in growth on glucose (data not shown). On the other hand, the upregulation of several genes (YIL011w/TIR3; YOR382w/FIT2; YOR383c/FIT3, and YER011w/TIR1) encoding glycosylphosphatidylinositol (GPI) anchored cell wall mannoproteins in strain L69, L62, and L60 could explain their higher content of mannans in cell wall (Table 1), whereas the downregulation of 3 out of the 5 Yapsin family genes shown to be involved in glucans homeostasis (Krysan et al., 2005) could account for lower β-glucans in these industrial yeasts as compared to lab strain. This transcriptomic change together with the slight upregulation (i.e., 1.5–2.5) of KTR6 implicated in mannosylation of proteins (Lussier et al., 1997), KEG1 involved in β-1,6 glucans synthesis (Nakamata et al., 2007), HLR1 and EMW1 encoding proteins implicated in cell wall maintenance and integrity (Versele and Thevelein, 2001) may contribute to a wall composition and structure of the industrial strains that differs from the laboratory strain. As a final note, we noticed a 1.6–2.2-fold increase in the expression of ERG1 (Table S2) in the 4 industrial yeast strains. The higher expression of this gene that codes for the first enzyme in ergosterol synthesis pathway from squalene (Leber et al., 1998) might be attributed to the tendency of these strains to be cultured under microaerobic conditions. To conclude, differences in cell wall composition and architecture between industrial and laboratory strains could be grossly inferred from differential genes expression. However, this transcriptomic analysis remains only informative to unravel the molecular cues that account for the biochemical and biophysical properties of cell wall and how these two parameters are interconnected.

The polysaccharides composition of the cell wall, Young modulus, adhesive events with concanavalin A, hydrophobicity and transcriptome are a set of biochemical, biophysics and molecular data that must be connected in one way or another. To unravel the connection between these high dimensional data where the number of variables (e.g., genes, proteins, metabolites) is much larger than the number of samples (i.e., yeast strains in this work), we employed the sparse methods of the mixOmics package. Exploration of relationships and correlation between these data sets requires reducing the dimension of the variables into components, from which can be retrieved correlations that are amenable to statistical inference about novel function or biological hypotheses by employing several sparse multivariate models (Gonzalez et al., 2009; Le Cao et al., 2009a). The sparse methods include a selection jointly performed with the search for correlated variables. We thus initiated our exploration analysis using the Sparse Partial least square (sPLS) regression mode which allowed to capture genes (component 1) whose the absolute expression level is the most correlated with the physico-chemical and biochemical variables (component 2). The correlation between component 1 and component 2 are projected into a correlation circle plot in which it can be seen that with, the exception of the variable “cell size,” all other parameters are located outside a 0.5 circle value, indicating either a positive (value > +0.5) or a negative (value < −0.5) correlation between them (Figure S6). According to Gonzalez et al. (2012), correlation between these variables is better visualized in a 2-dimensional heat map representation where relations between biophysical and biochemical variables are reported on the vertical axis whereas the horizontal axis shows the genes whose expression are the most correlated with these two variables (Figure 6). Confidence about this multivariate methodology sPLS is provided by retrieving predictable correlations such as between hydrophobicity and adhesion events (i.e., frequency of interaction of the concanavalin A-functionalized AFM tips on the yeast cell surface), as well as confirming that there is no simple correlation between the Young's modulus and any of the three types of cell wall polysaccharides (Dague et al., 2010; Schiavone et al., 2016). On the other hand, pertinent associations can be pointed out thanks to this statistical method such as a connection of β-1,3-glucans with the contour length or unexpected ones like the association of mannans and chitin as one would predict an association of chitin with β-1,6 glucans according to the cell wall structure proposed elsewhere (Klis, 1994).

On the horizontal axis, a set of more than 150 genes were retrieved from the 5 transcriptome data, the expression of which was the most correlated with either a physical or a biochemical variable. Detailed description of these genes with their assignated function as obtained from SGD database (https://www.yeastgenome.org/) is reported in Table 2. The sPLS regression analysis further distributed this list into 6 groups based on their closest association with either one of these variables. Group 1 which contained 23 genes was positively correlated to β-1,6-glucans and to chitin but negatively associated with the contour length and β-1,3-glucans. Surprisingly enough, this cluster was enriched of genes that encode sulfate assimilation pathway which is required for the biosynthesis of cysteine, an amino acid that is implicated in the cysteine-rich domain of cell wall sensors (Kock et al., 2015) and in disulfide link between CWP. Genes that belong to the ergosterol biosynthetic process were enriched in cluster 2 which was positively correlated with hydrophobicity, adhesion events and mannans, and weakly negatively associated with the Young's modulus. On the other hand, these same biochemical and biophysical variables were negatively correlated with genes in cluster 3 from which no particular GO biological or GO component process enrichment could be found. Nonetheless, we noticed that 30% of the genes in this cluster encode proteins with unknown function. The 18 genes that constitute cluster 4 were shown to correlate positively with the Young's modulus, cell size and β-1,6-glucans content. Main biological function identified in this cluster are genes encoding proteins involved in mitotic division and chromosome separation through tubulin and microtubulin polymerization/depolymerization, which is expected as regards to the cell size parameter. Group 5 which includes 25 genes showed positive correlation with contour length and β-1,3-glucans, and this correlation could be explained by the remarkable enrichment in the seripauperin multigenes (PAU) family encoding cell wall proteins. Finally, cluster 6 that contained about 30 genes showed anti-correlation with Young modulus, cell size and to a lesser extent β-1,6-glucans. However, there was no clear GO function enrichment in this cluster that could help us to explain this negative association.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.